Nvidia has announced a strategic partnership with Marvell to strengthen the networking infrastructure of so-called 'AI factories'.
This move is fundamentally about upgrading the 'plumbing' of massive data centers. As AI models grow larger, the communication speed between thousands of GPUs becomes a critical bottleneck. The solution lies in shifting from traditional pluggable optics to more advanced co-packaged optics (CPO) and silicon photonics. This technology transmits data using light instead of electricity, drastically reducing power consumption and increasing speed. Recognizing this, Nvidia recently invested a massive $4 billion in photonics suppliers Lumentum and Coherent to secure its supply chain, and this partnership with Marvell is the next logical step to integrate the technology.
The groundwork for this partnership was laid over the past year. First, in 2025, Nvidia unveiled its photonics-first roadmap, including the Spectrum-X networking platform and NVLink Fusion, which opened the door for collaboration with custom silicon partners like Marvell. Around the same time, Marvell was already showcasing its 1.6T optical technology and acquiring specialized firms like Celestial AI to bolster its 'Photonic Fabric' platform.
Second, events in March 2026 brought everything to a head. At its GTC conference, Nvidia heavily emphasized optical networking. Shortly after, Marvell demonstrated its comprehensive AI connectivity portfolio at the OFC conference and posted record data center revenues, proving market demand and its own readiness. The final piece was Nvidia's $4 billion investment, which signaled a serious commitment to building a robust U.S.-based photonics ecosystem, partly as a hedge against geopolitical supply chain risks.
Ultimately, this partnership represents a powerful convergence. It combines Nvidia’s architectural vision and immense capital with Marvell’s specialized silicon and photonics expertise. Together, they are building the high-speed, power-efficient interconnects essential for the next generation of AI supercomputers, ensuring the 'AI factories' of the future can operate at an unprecedented scale.
- Silicon Photonics: A technology that integrates optical components onto silicon chips, allowing data to be transmitted via light (photons) instead of electrical signals for higher speed and lower power consumption.
- Co-packaged Optics (CPO): An approach where optical connectivity components are placed on the same package as the main processor (like a GPU or switch), minimizing data travel distance to significantly improve power efficiency and performance.
- NVLink: A high-speed, direct GPU-to-GPU interconnect technology developed by Nvidia that allows for much faster data sharing between GPUs compared to traditional connections like PCIe.